Free · no account
What do the studies actually say?
546 findings on 58 everyday ingredients. Each names what a source measured, how sure anyone can be, and what is still a gap, with the source one tap away.
Or start from an example: magnesium creatine sleep
This is a library, not advice. No recommendation, no dose, no diagnosis, no ranking. We never ask about your symptoms, and nothing here is tailored to you.
How sure we can be about these
- High certainty, level 4 of 49
- Moderate certainty, level 3 of 4141
- Low certainty, level 2 of 4292
- Very low certainty, level 1 of 4104
Read from 313 sources: 311 PubMed records and 2 other documents. Every finding links to the one it came from.
- Things measured
- 136
- Questions covered
- 418
Search the findings
Search for a name you know: an ingredient like creatine, or something measured like sleep. Nothing you type is sent anywhere.
Five words, in plain language
- GRADE
- the standard method for rating how certain evidence is
- Confidence interval
- the range the true effect is likely to sit in
- Systematic review
- a study that pools all the trials on a question
- Risk of bias
- how likely a study’s design distorted its result
- Heterogeneity
- how much the pooled studies disagreed
Searching, filtering and sorting run in your own browser and need JavaScript. Every total above and below stays readable without it.
Where the library actually stands
The three numbers most places leave out.
- Questions we have read
- 409
- What our own product uses
- 31
- What changed after publication
- 45
of 645 on our to-read list
236 are still empty.
used when we answer a question
31 findings are used in answers; 16 of those carry a signed, machine-rechecked evidence chain. 2 reading-room-only findings are signed and independently replayed for public reading; they are not used by recommendation answers. In total, 515 are reading-room only.
An AI agent has re-read 511 of the reading-room findings against their saved sources, checking the quote, the numbers, the units and the direction. It confirmed 308 and could not confirm 203. No person read them; the owner authorised the agent to sign each batch with the owner’s key, and every build rechecks those signatures against these exact findings.
We have read 515 findings our product does not draw on. That says something about our product, not about the science.
findings carry a notice
across 26 sources. We show the notice beside the finding, and we never rewrite the number the source reported.
Another 9 questions we have published were never on our to-read list of 645, which makes 418 published in all.
Inside a finding, some things were reported by the source and some were not. How often each turns up across all 546 findings:
How to read a finding
Eight things worth knowing first.
- A source link is not evidence strength. Every finding links to its source, so you can check it. A weak study with a link is still a weak study.
- A fixed-rule rating is not an expert opinion. 515 of 546 ratings came from one rule applied to the source’s own fields, with every step shown and nobody signing them off.
- “Also used in OWP’s own product” describes us, not the science. The label says only whether we draw on a finding elsewhere. One we do not use is not weaker science.
- Not on our list is not absence of research. We work from our own to-read list of 645 questions. A question that is empty, or never made the list, is one we have not read yet.
- Higher or lower is not better or worse. Which way the numbers went reads “Higher”, “Lower”, “No clear difference” or “Mixed results” — a fact about the numbers, never a claim about anyone.
- A range crossing no-difference rules nothing out. Where a confidence interval crosses the no-difference point (1 for a ratio, 0 for a difference), the numbers fit a real effect and no effect equally well, so this finding cannot rule out no effect.
- A correction describes what happened to a source later. A notice is attached after publication. It never rewrites the number the source first reported.
- Ingredient names are not merged together. We keep names as recorded, so
psylliumandpsyllium_huskstay two entries and their findings are never pooled.
Behind each row are three kinds of information we keep apart: what the source reported, what we standardised, and what we worked out. How we work sets out the certainty rule in full.
What this page will not do
What we do not cover
This library shows what we have read so far, not every ingredient or everything measured. A question with no finding under it is not an answer, and nothing here says anything about any person, product or dose.
Gaps are written down, not hidden. Most findings come from published abstracts. 2 signed public-reading findings come from the full paper, and some come from other published documents. Each finding says what was read and names a missing detail for that reading: “Not in the abstract”, “Not in this document” or “Not recorded in this reading” — never a zero, never a blank and never a guess. A real zero is shown as a zero. Nothing here is a dose, a ranking or a “best” list.
Corrections and updates
Every finding shows one of three things: the exact wording of the notice, where the saved copy carries an erratum or an update; “We checked the saved copy: no notice”, where we read it and found none; or “No correction check recorded yet”, where nothing has been checked. We never call a source clear when nothing was checked.
Licence, version and privacy
All rights reserved — viewing only.
You may read this page and follow its links out to the sources. Copying, republishing or reusing extracts is not granted.
Library version
d9b91acf1decdd9e57277fccd4db2cce45a3bc5236ad15552e6f78aaa265e6bc, worked out from the library itself, so it changes whenever the library does.We don’t track you: this page sets no cookies and sends no analytics.